Enerelt Murakami, Eiji Yamada, and Erica Sioson

No. 181

January 2019

The Impact of Migration and Remittances on Labor Supply in Tajikistan

Study on Remittances and Household Finances in the Philippines and Tajikistan

Use and dissemination of this working paper is encouraged; however, the JICA Research Institute requests due acknowledgement and a copy of any publication for which this working paper has provided input. The views expressed in this paper are those of the author(s) and do not necessarily represent the official positions of either the JICA Research Institute or JICA. JICA Research Institute 10-5 Ichigaya Honmura-cho Shinjuku-ku Tokyo 162-8433 JAPAN TEL: +81-3-3269-3374 FAX: +81-3-3269-2054

1

The Impact of Migration and Remittances on Labor Supply in Tajikistan

Enerelt Murakami*, Eiji Yamada†, and Erica Sioson‡

Abstract

This paper examines the labor supply effects of migration and remittances in Tajikistan – a major labor migrant sending and remittance dependent country in Central Asia. We contribute to the literature in two ways. First, we effectively address the common methodological issues that result in biased estimates in analyses of migration and remittances. Our empirical work accounts for the endogeneity of migration and remittances with respect to the labor supply decisions of household members left at home, and for the self-selection of migrants and remittance senders through the application of a control function approach. Second, we apply our empirical model to unique high-frequency household panel data that further helps to remedy methodological problems present in cross-sectional studies. The findings suggest that having a migrant member and receiving remittances increases the reservation wages of the household members left at home, thereby reducing their labor supply and economic activity rate. This result is robust to different model specifications and definitions of migration and remittances. Keywords: Migration, remittances, labor market participation, economic activity rate, endogenous switching, Tajikistan * JICA Research Institute (Murakami.Enerelt@jica.go.jp) † JICA Research Institute and SciencesPo Paris ‡ Asian Development Bank. Institute and University of Tokyo. This paper has been prepared as part of a JICA Research Institute project entitled“Study on Remittances and Household Finances in the Philippines and Tajikistan.” We thank William Hutchins Seitz, Joao Pedro Wagner De Azevedo, and other colleagues in the World Bank for their academic as well as administrative efforts to arrange an institutional collaboration between the World Bank and the JICA Research Institute that allowed us to participate in the survey and access the data from the L2TJK project.

2

1. Introduction

Tajikistan has experienced unprecedented out-migration since achieving independence in 1991.

In the early years of its independence, migration was mostly driven by cultural and ethnic

motivations triggered by the collapse of the Soviet Union and the subsequent civil war. More

recently, however, economic reasons are driving migration, Tajiks are seeking better job

opportunities abroad to improve their earning potential. With productivity, growth, and job

creation not catching up with rapid population growth, only about a half of the total working-age

population was registered employed in Tajikistan in 2017 (World Bank 2017). The lack of job

opportunities at home drives many Tajiks to seek employment abroad, with a popular destination

being Russia due to historical and cultural connections. A recent nationally representative

household survey conducted by the World Bank and the German Federal Enterprise for

International Cooperation (GIZ) (2013) shows that almost 40 percent of households have at least

one member abroad in work, of which about 90 percent are in Russia.

The contribution of labor migrants to the Tajikistan economy is enormous. Remittances

received from labor migrants have constituted 30 to 50 percent of the country’s GDP since 2006

(World Bank 2017), making Tajikistan one of the most remittance-dependent countries in the

world. The remittance flows provide the most important source of external funds in the country,

surpassing foreign direct investment (FDI) and official development assistance (ODA) flows by

more than ten times. At the macro level, the remittances sent by labor migrants substantially

contribute to GDP growth and poverty reduction but create an excessive dependency on the

economies of remittance-source countries. Most research on migration and remittances in

Tajikistan focuses on their impacts on economic growth and poverty reduction at the macro level.

However, understanding how these large migrant and remittance flows affect migrants’

households and their economic behavior in the home country is also important if the

3

Government is to reduce Tajikistan’s excessive dependence on remittances and exposure to

external turbulence.

Migration and remittances can have various impacts on the labor market decisions of the

left-behind household members (Démurger 2015). First, migration through remittances may

increase the reservation wage of non-migrant household members and thereby reduce their labor

supply in the local economy. Second, remittances may lift the liquidity constraints faced by

migrant households and create more opportunities for non-migrant household members in

productive entrepreneurship activities. Third, loss of the income contribution of a migrant

household member in the short-run could lead to a non-migrant member who was previously not

engaged in paid employment seeking employment to replace the lost income.

The interest in labor market participation derives in general from its bearing on

long-term economic growth. With migration and remittances figuring substantially in the

development discourse in Tajikistan, the need to determine whether migration and remittances

have a positive, or detrimental, impact on long-term economic growth becomes more urgent.

The labor supply effect of migration and remittances is particularly important for countries like

Tajikistan. Nevertheless, the global evidence is not conclusive on whether international

migration and remittances affect the labor supply of the left-behinds positively or negatively.

Past empirical studies on Tajikistan are limited, with most of them suggesting that the

reservation wage effect discourages the families of migrants from working. For example,

Abdulloev (2013) finds that satisfaction regarding jobs offered in Tajikistan is significantly

lower for families who have international migrants. Using the Tajikistan Living Standards

Survey (TLSS) conducted in 2003, Justino and Shemyakina (2012) find a negative effect on

household labor supply. However, these studies used cross-sectional data which are not ideal

when dealing with the issue of endogeneity of migration and remittances. Furthermore, these

studies used data collected in the early 2000s and do not capture the situation after Russia’s

4

economic decline in 2014 and the resulting stricter regulations on immigrants implemented in

2015.

The objective of this paper is therefore to contribute to the scarce empirical literature on

the impact of migration on non-migrant labor supply in Tajikistan. The contribution of the paper

is twofold. First, we analyze the latest data collected through the ongoing “Listening to

Tajikistan (L2TJK)” project that is being conducted by the World Bank. Employing a Telephone

Assisted Personal Interview (TAPI) technique, the L2TJK collects the socio-economic data of

800 households every two weeks. As of November 2017, the project had collected 32 rounds of

high frequency panel data. At each round, the data show that more than 30 percent of households

have at least one migrant member on average. Since the interviews are conducted frequently, the

data allow us to detect the instantaneous responses of households to various shocks without

severe recall errors. To the best of our knowledge, there has been no study which uses high

frequency panel data similar to the L2TJK on this topic. Second, our empirical strategy

addresses the common methodological issues – endogeneity and selection bias – present in

studies of migration and remittances, by applying a control function approach based on

Murtazashvili and Wooldridge (2016). When estimating the impact of remittances and migration

on the labor market participation of household members left behind, we need to consider the

possibility of endogeneity, simultaneity, and self-selection. Decisions on international migration,

remittances, and domestic labor market participation are likely to be made simultaneously or

cause each other. Moreover, migrants and remitters are not drawn from a randomly selected

sample population, but from individuals who self-select into these activities. The advantage of

the Murtazashvili and Wooldridge (2016) approach is that not only does it correct for selection

bias and endogeneity, it also is less restrictive and less computationally expensive compared to

competing models.

Our results show a large reservation wage effect. On average, if a household sends a

migrant, or receives remittances, the labor market participation rate of the left-behind members

5

declines by 8 and 11 percentage points respectively. This is higher than the estimates by Justino

and Shemyakina (2012) using a similar definition, whose results range from 5 to 8 percent. With

international migration becoming a familiar and sometimes preferred occupational choice for

many Tajiks, the reservation wage effect can be detrimental to the nation’s growth potential in

the long run, through the slowed development of the domestic labor market and lower human

capital accumulation. This negative effect of migration and remittances has become a concern

for policy makers who try to enhance domestic job opportunities. Job creation remains a

daunting task, as found out by one of the authors of this paper after interviewing some high

officials at the Tajikistan’s Ministry of Labor. Domestic jobs continue to be unattractive, as

wages remain low. With migration to Russia as a familiar occupational option, which provides

better wages, people are not willing to work at the domestic wage level.

The rest of the paper is organized as follows: Section 2 discusses the recent patterns of

migration, remittances, and the labor market in Tajikistan. Section 3 reviews the related

literature on the impacts of remittances on labor supply. Sections 4 and 5 describe the

methodology and data employed in the analysis. Section 6 presents and discusses results.

Section 7 performs robustness checks for different specifications, and Section 8 concludes the

paper.

2. Migration, remittances, and domestic labor market patterns in Tajikistan

In the neoclassical theory of migration, the reasons why people migrate are often categorized

into push and pull factors that are related to the economic context of the flow of labor (Kurekova

2011). These factors pertain to the relational drivers of migration, both from the migrant-sending

country perspective (push), and from the migrant-receiving country (pull). Both push and pull

factors coincide to make Tajikistan one of the biggest exporters of labor in the region. Large

wage differentials between Tajikistan and Russia as well as other destination countries are often

6

cited as major pull factors, while a shortage of jobs relative to the population growth and low

wages in the former are considered as the main push factors.

Figure 1 shows the distribution of migrant Tajiks in different parts of the world. Data

from the United Nations Population Division, Department of Economic and Social Affairs (UN

DESA 2017) shows that Europe hosts the highest number of Tajiks with about 92.51 percent of

the migrant Tajik population being in Europe in 2017, a considerable increase from the 87.53

percent in 1990. Most of these migrants are in Russia, and the majority are involved in itinerant

jobs from spring to fall, often in the construction industries (Erlich 2006). In 2013, the World

Bank and GIZ found that about 40 percent of households in Tajikistan have at least one migrant

member.

Figure 1. Stock of migrant Tajiks in the world by continent, 1990-2017

Source: UN DESA 2017.

The growth of migrant workers shows a consistent number of the Tajik population

engaged in migratory flows. By 1999, according to the TLSS, about 1.5 percent of households

had migrant workers. This grew to 5 percent in 2007. By 2008, the total number of labor

migrants according to the Ministry of Labor was 805,000, compared to 224,000 in 2003

050,000

100,000150,000200,000250,000300,000350,000400,000450,000500,000550,000600,000

1990 1995 2000 2005 2010 2015 2017

Asia Europe Others

7

(Olimova 2010). Brown, Olimova and Boboev (2008) find that 37.3 percent of households had

at least one migrant member and that about 700,000 people, about 500,000 of whom were

working in Russia, were considered temporary migrants.

By the end of 2008 and beginning of 2009, the number of labor migrants had declined by

a fourth, as many Tajik migrants returned to Tajikistan largely because of the global financial

crisis of 2008 and the sharp decline in economic activity within Russia. Before this, migratory

flows from Tajikistan to Russia had heightened during the breakup of the Soviet Union and the

ensuing civil war in Tajikistan. The civil war displaced as much as 20 percent of the country’s

population (Yormirzoev 2017), becoming the main push factor for migration. The succeeding

years saw a combination of job shortages, demographic pressures and limited land area begin to

push Tajiks to migrate and work, primarily in Russia. While some Tajiks migrate to other

countries to work, Russia remains the main destination for the majority. This results in Tajikistan

being dependent, hence vulnerable, to changes in the Russian economy.

Figure 2. Arrived Tajiks in Russia, 1997-2016

Source: Russian Federation Federal State Statistics Services (2017).

0

10000

20000

30000

40000

50000

60000

8

Figure 2 shows a steady decrease in the number of arriving Tajiks in Russia from 2000 to

2005. From more than 11,000 in 2000, it dropped more than half in 2005. This is most probably

attributable to the migration policy reform that was implemented to simplify the arrivals and

settlements of migrants which affected temporarily employed migrants from the Commonwealth

of Independent States (CIS) member countries, including Tajikistan (Mukomel 2014). By

2007-2008, Russia was implementing liberal migration reform, resulting in a consistent increase

in arrivals, reflected in the steady arrivals of Tajiks until 2008, when the numbers plummeted in

2010 following the Global Financial Crisis. Subsequently, migration rates rebounded and

steadily increased until 2014. From 54,658 entrants in 2014, the numbers however dropped to

47,638 in 2015. This reflects the revision of the visa-granting procedures implemented in 2015

by Russia. However, by 2016, there was a relative increase in numbers as arrived Tajiks

increased to 52,676. This pattern is also reflected in Figure 3 showing a decline in the number of

working Tajiks in Russia by 2015.

Figure 3. Number of Tajiks with work permits in Russia (in thousands)

Source: Russian Federation Federal State Statistics Services (2016).

0

200

400

600

800

1000

1200

9

The aggregate flow of remittances to Tajikistan is volatile (Figure 4) just like the number

of Tajik migrants which is largely affected by the migration policy of the Russian government.

Migrants’ remittances account for a significant portion of the small country’s GDP making it

extremely vulnerable to changes in the Russian economy. Migrant remittance inflows to

Tajikistan experienced a steady increase from 2002 to 2008, almost doubling each year until

2009. In 2009, remittances fell to 1.748 billion from 2.544 billion in 2008, reflecting the impact

of the Global Financial Crisis felt through its impact on Russia. By 2010, remittance inflows to

the country had recovered, accounting for 31 percent of the country’s GDP or about 2.254 billion

US Dollars.

By 2011, migrant remittances inflows amounted to 2.68 billion US Dollars (41.7 percent

of GDP) and steadily increased until they reached a peak in 2013 at 4.219 billion US Dollars

(43.5 percent of GDP). The share of remittance to the country’s GDP dropped to 36.6 percent in

2014 after the devaluation of the Russian ruble. This is also backed by the data in Figure 3 where

the numbers of Tajik workers in Russia declined in the same year. The decline in remittances has

been consistent since 2013 with a relative increase in 2016, though the share of remittances to

GDP continued to drop to 26.9 percent in that year. By 2017 migrant remittance inflows were

reduced by more than half at 2.031 billion US Dollars.

10

Figure 4. Remittance inflows to Tajikistan, 2002-2017 (US$ million)

Source: World Bank (2017).

Another important factor to consider driving Tajiks to migrate to Russia is the lack of

employment opportunities in the country. After the civil war in the 1990s, some forms of

employment in the country vanished. A sharp rise in unemployment due to the closure of many

enterprises together with high birth rates became strong push factors in the wake of the Soviet

Union’s collapse (Olimova 2010). Employment in industrial production declined from 21

percent in 1991 to eight percent in 2003. Most shifted to agriculture, retail trade, services, and

household production while some stopped looking for jobs completely (Olimova 2010). Since

1994-1995, on the other hand, Russia’s sustained economic growth, comparatively high wages,

ease of migration, and labor shortages have pulled Tajiks towards migrating (Erlich 2006,

Olimova 2010). Tajiks looking for a job elsewhere have found jobs in Russia, filling a demand

for unskilled labor in that country.

So, a combination of high unemployment and high population growth have served as

push factors driving people out of the country to work in Russia. Figure 5 shows the total

population, vis-à-vis total labor force and the economically active population from 1990 to 2016.

These data show that as the population and the number of people considered as part of total labor

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

11

resources constantly increase, the number of people who are economically active does not. The

increase in the economically active population tends to be slow and in certain years even

decrease, but high birth rates have contributed to a rapid and continuous increase in population.

The current median age is around 22 years, suggesting a young population with a potentially

large labor force. However, as noted, while the potential labor force has increased because of

natural population growth, the population engaged actively in the economy has not increased as

fast. Looking at the registered unemployed we see that by the first quarter of 2016, 54,000 Tajiks

were registered unemployed, a drop from the 57,000 registered in the first quarter of 2015.

However, this is not the complete story as many unemployed Tajiks remain unregistered at the

unemployment office.

Figure 5. Population, labor force and economically active, 1990-2016 (in thousands)

Source: Statistical Agency of Tajikistan. (2017), an agency under the President of the Republic of Tajikistan. Retrieved 9, 1, 2017, from http://www.stat.tj/en/.

In a report by the European Union in 2010, job creation was highlighted as among

significant areas of concern (EU 2010). No inventory on job creation is available however and it

becomes impossible to assess the opportunities for employment growth (EU 2010). According

0100020003000400050006000700080009000

10000

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

Population Total labor force Economically active

12

to a report by Strokova and Ajwad (2017) for the World Bank, the potential workforce however

is growing but it remains underutilized, most probably attributable to the slow growth of job

creation. The majority of those working are employed in the informal sector in low quality jobs.

More than 60 percent of the total employment in the country is employed in agriculture and

related sectors (Strokova and Ajwad 2017). The service sectors employ about 30 percent of the

employed population, more than the population employed by the industry sector at less than 20

percent. One possible reason is that firms in the private sector remain small and young (Strokova

and Ajwad 2017). In general, Strokova and Ajwad (2017) note that “labor has moved out of the

more productive sectors, such as industry, into low-productivity services and agriculture sectors,

where domestic job creation is the highest.”

We can see from the data presented above that a combination of push and pull factors

contribute to many Tajiks leaving for other countries to find work. A couple of conclusions can

be drawn: (1) high birth rates combined with slow job creation, especially in more productive

sectors drive people to look for jobs elsewhere; (2) a combination of the historical relations

between Russia and Tajikistan as well as the higher wages in the former make Russia a preferred

destination for many Tajik migrants; and (3) the number of Tajiks migrating to work in Russia

and most importantly the amount of remittances sent back to Tajikistan remains volatile, affected

by changes in Russia’s migration policies and its economy.

3. Literature review

The impacts of migration and remittances on other development indicators such as consumption,

immediate well-being, increases in per-capita income, and on compensation for negative shocks

have been substantiated in the literature (see Ratha 2013; Acosta, Calderon, Fajnzylber and

López 2008; Hildebrandt and McKenzie 2005), while the long-term effects of migration and

remittances especially on productivity remain inconclusive. This inconclusiveness has been

13

attributed to several methodological issues such as selection bias, reverse causality, and omitted

variable bias (Adams 2011). Given this, we identify three strands of literature that discuss the

impacts of remittances on labor supply: first, that remittances can decrease participation in the

labor market; second, that remittances have no effect on the labor supply; and third, that

remittances can increase liquidity allowing households to invest in human capital.

Following the neo-classical model of labor supply, it is assumed that individuals allocate

time to both market and non-market activities. According to this perspective, the decision in

allocating time to these activities is determined by a number of factors such as wage and

non-labor income (Cox-Edwards and Rodriguez-Oreggia 2009). Labor-leisure theory notes that

remittances if considered as non-labor income can decrease the propensity of non-migrant

household members to participate in the labor market. Receipts of remittances can thus increase

the reservation wage of members left in the household. Studies such as Acosta (2007), Acosta,

Lartey, and Mandelman (2009), Chami, Fullenkamp, and Jahjah (2005) and Chami, Hakura, and

Montiel (2012) also contribute to the evidence that remittances can have negative effects on

labor supply and hours worked by members left in the home country. Kim (2007) in a study on

Jamaica using fixed-effect regression looked at the factors that drive a wedge between

productivity and reservation wages and note that recipient household heads regardless of gender

tend to work fewer hours than non-recipient heads. One criticism of the Kim (2007) study,

however, is that it did not control for selection in the receipt of remittances (Adams 2011).

Chami, Fullenkamp and Jahjah (2005) note the negative effect of remittances on growth

and productivity. Chami, Hakura and Montiel (2012) conclude that positive technological

shocks can induce labor supply through an increase in real wages, and that remittances in

response can contract, reducing demand for leisure over labor, which in effect would increase the

labor supply. Acosta (2007), in a study using a two-stage least-squares model and instrumental

variable approach on a four-year panel survey in El Salvador, highlighted the importance in

14

looking at groups and noted that women are more likely to quit the labor market than men, but

that both men and women do reduce hours worked when their households receive remittances.

While the perspective that remittances can reduce the labor supply has dominated the

literature, it has not been unchallenged. If remittances are considered as labor income and as

income that otherwise the migrant member would contribute to the household if he or she has not

left the country, then there should be no effect on labor supply. Jansen, Vacaflores and Naufal

(2012) indicate that if remittances are not just a gift from relatives, nor additional non-labor

income, but are in fact a household decision regarding labor allocation, then these inflows may

not have such huge impacts on a household’s domestic work effort. In this vein, studies such as

Assaad (2011), Cabegin (2006), Cox-Edwards and Rodriguez-Oreggia (2009), and Funkhouser

(1992) argue that there is no effect.

Another interesting finding is that while involvement in the formal labor market has

decreased among remittance-receiving households, involvement in the informal sector has

increased suggesting that household members of remittance-receiving households tend to favor

work that provides more mobility and flexibility. This is supported by the Funkhouser (1992)

study using fixed effects that showed that an increase in remittances would have a negative

impact on the labor force participation of members left in the household but would also have a

positive impact on self-employment. This study however was not without shortcomings. Adams

(2011) note that like the Kim (2007) study, Funkhouser (1992) did not control for selection in the

receipt of remittances and therefore the results could be biased.

Cabegin (2006), on the other hand, used the two-stage probit-OLS method in a study on

remittance-receiving households in the Philippines. In a study that corrected the biases in the

Funkhouser (1992) and Kim (2007) studies, she noted that for married couples, the participation

in migration abroad of one partner can change the labor participation and supply of the other

partner. Her findings are somewhat similar to Acosta (2007) though she argued that this operates

differently for men and women. She further found that having school-aged children can reduce

15

market participation for married women in respect to full-time paid employment. The effects

however are limited for married men, though the results suggest that an increase in a migrant

wife’s remittances can reduce the likelihood of non-employment for men. The resulting change

in the labor supply can be a function of the change in roles household members assume upon the

out-migration of the migrant member.

This is especially true if it is the household head that migrates. In a 2007 study on the

labor market inactivity of migrant-sending households in Moldova, Görlich, Mahmoud and

Trebesch (2007) offer a new perspective on understanding labor-leisure theory as it applies to

remittances and migration. They argue that the inactivity in the labor market of

remittance-receiving households is not because they consume more leisure, but because of

intra-household labor substitution and increased university enrolment. This was further

elaborated by Assaad (2011) in a study on the labor supply responses of women left behind in

Egypt. Using cross-sectional data from the Egypt Labor Market Panel Survey of 2006, this study

found that while a male household member’s migration can induce women in rural Egypt to

respond to this migration by increasing their labor supply, women are more likely to engage in

unpaid family work. The reason for this is that the a household member’s migration ultimately

meant loss of labor for the remaining members, and women are expected to replace this labor.

Finally, another group of studies point to the positive impacts of remittances in terms of

the increase in liquidity allowing households to invest in human capital, and to some extent

financial capital. Calero, Bedi, and Sparrow (2009), using data from Ecuador, note that

remittances can facilitate human capital investments. They used data on availability of bank

offices in source countries as instruments to understand whether remittances can increase school

enrollments. They also found out that remittances are being used to fund education when

households are faced with economic shocks. Nsiah (2010) also came to a similar conclusion, on

a much larger scale. In their study on remittances and growth in Africa, they found that

16

remittances can provide alternative ways to finance investments thereby overcoming liquidity

constraints.

This study on the impact of migration and remittances on the labor supply of left-behind

household members aims to contribute to this debate. It is a response to the call for empirical

evidence on the impacts of migration and remittances in Tajikistan. As mentioned, the

importance of labor market participation lies on its bearing on long-term economic growth. With

many Tajiks migrating and with remittances constituting substantial shares in the country’s GDP,

the need to determine whether migration and remittances have a positive, or detrimental, impact

on long-term economic growth becomes more urgent.

4. Methodology

Estimating the impact of remittances or migration on the labor market participation of household

members left behind needs to consider the possibility of endogeneity and self-selection. In other

words, remittances and migration are endogenous to the labor market participation of household

members left behind, because a household can self-select into the status of sending migrants or

receiving remittances by an unobserved cause, which simultaneously affects their labor market

participation at home. Furthermore, a shock to the labor supply of left behind members (e.g., job

losses) can be a direct cause of the migration/remittance decision. Because of these omitted

variables and the impact of reverse causality, simply regressing the labor market participation on

migration/remittance status by OLS will deliver a biased estimate.

In the presence of endogeneity and self-selection bias, the estimation must consider the

unobserved heterogeneity that simultaneously affects remittances/migration and labor market

participation decisions. To correct for this endogeneity and selection bias, we employ a control

function approach to estimate an endogenous switching regression model for panel data

following Murtazashvili and Wooldridge (2016). This model is a type of two-stage estimation

17

where the first stage (the regime switching equation) estimates the determinants of the regime

switching variable, and the impact of the regime switching variable on the outcome is estimated

at the second stage (the outcome equation). We use a traditional endogenous switching

regression model that allows different coefficients across two different regimes in the outcome

equation of the following form:

𝑦𝑦𝑖𝑖𝑖𝑖1(𝓂𝓂) = 𝑥𝑥𝑖𝑖𝑖𝑖1𝛽𝛽𝑚𝑚 + 𝑐𝑐𝑖𝑖1𝑚𝑚 + 𝑢𝑢𝑖𝑖𝑖𝑖1𝑚𝑚, 𝑡𝑡 = 1,⋯ ,𝑇𝑇 and 𝑚𝑚 = 0,1 (1)

Where: 𝑦𝑦𝑖𝑖𝑖𝑖1(𝑚𝑚) is the outcome variable, the labor market participation of members left behind, of

household 𝑖𝑖 in round 𝑡𝑡 specific to the regime 𝑚𝑚 = 0,1.

In our case, 𝑚𝑚 = 1 indicates that the household has (a) migrant(s) or is receiving

remittances, and 𝑚𝑚 = 0 means otherwise; 𝑥𝑥𝑖𝑖𝑖𝑖1 is the vector of explanatory variables; 𝑐𝑐𝑖𝑖1𝑚𝑚 is

the household-specific unobserved heterogeneity in the regime 𝑚𝑚; 𝑢𝑢𝑖𝑖𝑖𝑖1𝑚𝑚 is an idiosyncratic

error term; and 𝑇𝑇 is the number of rounds in the panel data.1 We assume that the explanatory

variables 𝑥𝑥𝑖𝑖𝑖𝑖1 may contain continuous endogenous explanatory variables (EEVs) and strictly

exogenous explanatory variables 𝑧𝑧𝑖𝑖𝑖𝑖1 with respect to the idiosyncratic error, 𝑢𝑢𝑖𝑖𝑖𝑖1𝑚𝑚 . In our

model, 𝑥𝑥𝑖𝑖𝑖𝑖1 contains only the strictly exogenous variables 𝑧𝑧𝑖𝑖𝑖𝑖1, which are uncorrelated with

𝑢𝑢𝑖𝑖𝑖𝑖1𝑚𝑚.

Let 𝑦𝑦𝑖𝑖𝑖𝑖2 denote our regime-switching endogenous binary variable, taking the value 1 if

a household has (a) migrants or receives remittances and 0 otherwise. By substituting the

following counterfactual framework:

�𝑦𝑦𝑖𝑖𝑖𝑖1

(0) = 𝑥𝑥𝑖𝑖𝑖𝑖1𝛽𝛽0 + 𝑐𝑐𝑖𝑖10 + 𝑢𝑢𝑖𝑖𝑖𝑖10𝑦𝑦𝑖𝑖𝑖𝑖1

(1) = 𝑥𝑥𝑖𝑖𝑖𝑖1𝛽𝛽1 + 𝑐𝑐𝑖𝑖11 + 𝑢𝑢𝑖𝑖𝑖𝑖11

1 As explained in Section 5 below, our data consists of 𝑇𝑇 = 32 rounds.

18

into a generic expression for 𝑦𝑦𝑖𝑖𝑖𝑖1(𝑚𝑚); 𝑦𝑦𝑖𝑖𝑖𝑖1 = (1 − 𝑦𝑦𝑖𝑖𝑖𝑖2)𝑦𝑦𝑖𝑖𝑖𝑖1

(0) + 𝑦𝑦𝑖𝑖𝑖𝑖2𝑦𝑦𝑖𝑖𝑖𝑖1(1), we can rewrite the regime

dependent outcome equation (1) into a switching regression equation with constant coefficients

as:

𝑦𝑦𝑖𝑖𝑖𝑖1 = 𝑥𝑥𝑖𝑖𝑖𝑖1𝛽𝛽0 + 𝑦𝑦𝑖𝑖𝑖𝑖2𝑥𝑥𝑖𝑖𝑖𝑖1(𝛽𝛽1 − 𝛽𝛽0) + 𝑐𝑐𝑖𝑖10 + 𝑦𝑦𝑖𝑖𝑖𝑖2(𝑐𝑐𝑖𝑖11 − 𝑐𝑐𝑖𝑖10) + 𝑢𝑢𝑖𝑖𝑖𝑖10 + 𝑦𝑦𝑖𝑖𝑖𝑖2(𝑢𝑢𝑖𝑖𝑖𝑖11 − 𝑢𝑢𝑖𝑖𝑖𝑖10)

= 𝑥𝑥𝑖𝑖𝑖𝑖1𝛽𝛽0 + 𝑦𝑦𝑖𝑖𝑖𝑖2𝑥𝑥𝑖𝑖𝑖𝑖1𝛾𝛾1 + 𝑐𝑐𝑖𝑖10 + 𝑦𝑦𝑖𝑖𝑖𝑖2(𝑐𝑐𝑖𝑖11 − 𝑐𝑐𝑖𝑖10) + 𝑢𝑢𝑖𝑖𝑖𝑖10 + 𝑦𝑦𝑖𝑖𝑖𝑖2(𝑢𝑢𝑖𝑖𝑖𝑖11 − 𝑢𝑢𝑖𝑖𝑖𝑖10) (2)

In equation (2), the regime-switching variable, 𝑦𝑦𝑖𝑖𝑖𝑖2 , interacts with both time-invariant and

time-varying observable and unobservable variables. Following Murtazashvili and Wooldridge

(2016), we allow for the correlation between unobservables and the strictly exogenous

explanatory variables by applying the Mundlak (1978) device.

Let 𝑒𝑒𝑖𝑖𝑖𝑖0 ≡ 𝑐𝑐𝑖𝑖10 + 𝑢𝑢𝑖𝑖𝑖𝑖10, 𝑒𝑒𝑖𝑖𝑖𝑖1 ≡ 𝑐𝑐𝑖𝑖11 + 𝑢𝑢𝑖𝑖𝑖𝑖11, and 𝑣𝑣𝑖𝑖𝑖𝑖1 ≡ 𝑒𝑒𝑖𝑖𝑖𝑖1 − 𝑒𝑒𝑖𝑖𝑖𝑖0 and assume that

𝑒𝑒𝑖𝑖𝑖𝑖0 = 𝑧𝑧�̅�𝑖𝜌𝜌0 + 𝜂𝜂𝑖𝑖𝑖𝑖0 (3)

𝑣𝑣𝑖𝑖𝑖𝑖1 = 𝑧𝑧�̅�𝑖𝜌𝜌1 + 𝜂𝜂𝑖𝑖𝑖𝑖1 (4)

Where: 𝑧𝑧�̅�𝑖 = 1𝑇𝑇∑ 𝑧𝑧𝑖𝑖𝑖𝑖𝑇𝑇𝑖𝑖=1 , and (𝜂𝜂𝑖𝑖𝑖𝑖0, 𝜂𝜂𝑖𝑖𝑖𝑖1) are assumed to be independent of 𝑧𝑧𝑖𝑖𝑖𝑖.

The assumptions (3) and (4) impose strict exogeneity of 𝑧𝑧𝑖𝑖𝑖𝑖 with respect to the

idiosyncratic errors (𝑢𝑢𝑖𝑖𝑖𝑖11,𝑢𝑢𝑖𝑖𝑖𝑖10). Substituting (3) and (4) into (2) gives:

𝑦𝑦𝑖𝑖𝑖𝑖1 = 𝑥𝑥𝑖𝑖𝑖𝑖1𝛽𝛽0 + 𝑦𝑦𝑖𝑖𝑖𝑖2𝑥𝑥𝑖𝑖𝑖𝑖1𝛾𝛾1 + 𝑧𝑧�̅�𝑖𝜌𝜌0 + 𝑦𝑦𝑖𝑖𝑖𝑖2𝑧𝑧�̅�𝑖𝜌𝜌1 + 𝜂𝜂𝑖𝑖𝑖𝑖0 + 𝜂𝜂𝑖𝑖𝑖𝑖1 (5)

The binary response correlated random effects model for 𝑦𝑦𝑖𝑖𝑖𝑖2 is given by:

𝑦𝑦𝑖𝑖𝑖𝑖2 = 1[𝜅𝜅𝑖𝑖2 + 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋2 + 𝑧𝑧�̅�𝑖𝛿𝛿2 + 𝜈𝜈𝑖𝑖𝑖𝑖2 > 0] (6)

Where: 1[⋅] is an indicator function, 𝜅𝜅𝑖𝑖2 is a round fixed effect, and:

19

(𝜂𝜂𝑖𝑖𝑖𝑖0,𝜂𝜂𝑖𝑖𝑖𝑖1, 𝑣𝑣𝑖𝑖𝑖𝑖2) are assumed to be independent of 𝑧𝑧𝑖𝑖𝑖𝑖, and 𝜈𝜈𝑖𝑖𝑖𝑖2~𝑁𝑁(0,1) (7)

The error term 𝜈𝜈𝑖𝑖𝑖𝑖2 is allowed to have serial correlation within the panel (𝑖𝑖). We are

interested in estimating the structural equations of a household’s labor force participation rate

and economic activity rate as given by Equation (1) with a reduced form selection equation for a

migration state or a remittance receipt state given by Equation (6). Under the assumption (7), we

can write:

E(𝑣𝑣𝑖𝑖𝑖𝑖2|𝑦𝑦𝑖𝑖𝑖𝑖2, 𝑧𝑧𝑖𝑖𝑖𝑖) = ℎ(𝑦𝑦𝑖𝑖𝑖𝑖2, 𝜅𝜅𝑖𝑖2 + 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋2 + 𝑧𝑧�̅�𝑖𝛿𝛿2) (8)

Where: ℎ(∙) is the generalized error function,2 determined by:

ℎ(𝑦𝑦𝑖𝑖𝑖𝑖2,𝜅𝜅𝑖𝑖2 + 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋2 + 𝑧𝑧�̅�𝑖𝛿𝛿2) = 𝑦𝑦𝑖𝑖𝑖𝑖2𝜆𝜆(𝜅𝜅𝑖𝑖2 + 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋2 + 𝑧𝑧�̅�𝑖𝛿𝛿2)

−(1 − 𝑦𝑦𝑖𝑖𝑖𝑖2)𝜆𝜆(− 𝜅𝜅𝑖𝑖2 − 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋2 − 𝑧𝑧�̅�𝑖𝛿𝛿2) (9)

Where: 𝜆𝜆(𝜅𝜅𝑖𝑖2 + 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋2 + 𝑧𝑧�̅�𝑖𝛿𝛿2) is the inverse Mills ratio (IMR). Then, we assume that:

E(𝜂𝜂𝑖𝑖𝑖𝑖0|𝑣𝑣𝑖𝑖𝑖𝑖2) = 𝜉𝜉0𝑣𝑣𝑖𝑖𝑖𝑖2 and E(𝜂𝜂𝑖𝑖𝑖𝑖1|𝑣𝑣𝑖𝑖𝑖𝑖2) = 𝜉𝜉1𝑣𝑣𝑖𝑖𝑖𝑖2 (10)

Where: 𝜉𝜉0 = 0 and 𝜉𝜉1 = 0 imply that selection is exogenous. By iterated expectations:

E(𝜂𝜂𝑖𝑖𝑖𝑖0 + 𝑦𝑦𝑖𝑖𝑖𝑖2𝜂𝜂𝑖𝑖𝑖𝑖1|𝑦𝑦𝑖𝑖𝑖𝑖2, 𝑧𝑧𝑖𝑖𝑖𝑖) =

𝜉𝜉0ℎ𝑖𝑖𝑖𝑖2(𝑦𝑦𝑖𝑖𝑖𝑖2,𝜅𝜅𝑖𝑖2 + 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋2 + 𝑧𝑧�̅�𝑖𝛿𝛿2) + 𝜉𝜉1𝑦𝑦𝑖𝑖𝑖𝑖2ℎ𝑖𝑖𝑖𝑖2(𝑦𝑦𝑖𝑖𝑖𝑖2,𝜅𝜅𝑖𝑖2 + 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋2 + 𝑧𝑧�̅�𝑖𝛿𝛿2) (11)

2 See Murtazashvili and Wooldridge (2016) for the definition.

20

These generalized error terms are added into Equation (5) to correct for the endogeneity

of the regime switch variable, 𝑦𝑦𝑖𝑖𝑖𝑖2 (migrate/not migrate or remit/not remit). Then Equation (5)

becomes:

𝑦𝑦𝑖𝑖𝑖𝑖1 = 𝑥𝑥𝑖𝑖𝑖𝑖1𝛽𝛽0 + 𝑦𝑦𝑖𝑖𝑖𝑖2𝑥𝑥𝑖𝑖𝑖𝑖1𝛾𝛾1 + 𝑧𝑧�̅�𝑖𝜌𝜌0 + 𝑦𝑦𝑖𝑖𝑖𝑖2𝑧𝑧�̅�𝑖𝜌𝜌1 + 𝜉𝜉0ℎ𝑖𝑖𝑖𝑖2(𝑦𝑦𝑖𝑖𝑖𝑖2,𝜅𝜅𝑖𝑖2 + 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋2 + 𝑧𝑧�̅�𝑖𝛿𝛿2)

+𝜉𝜉1𝑦𝑦𝑖𝑖𝑖𝑖2ℎ𝑖𝑖𝑖𝑖2(𝑦𝑦𝑖𝑖𝑖𝑖2,𝜅𝜅𝑖𝑖2 + 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋2 + 𝑧𝑧�̅�𝑖𝛿𝛿2) + 𝜔𝜔𝑖𝑖𝑖𝑖 (12)

E�(𝜔𝜔𝑖𝑖𝑖𝑖)�𝑦𝑦𝑖𝑖𝑖𝑖2, 𝑧𝑧𝑖𝑖𝑖𝑖� = 0 (13)

Where: 𝜔𝜔𝑖𝑖𝑖𝑖 is the implied error in Equation (11).

To consistently estimate the coefficient of (12), we follow a two-step procedure as

proposed by Murtazashvili and Wooldridge (2016). In the first stage, we estimate a pooled probit

model of the following form for the selection equation (6):

P(𝑦𝑦𝑖𝑖𝑖𝑖2 = 1) = Φ(𝜅𝜅𝑖𝑖2 + 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋2 + 𝑧𝑧�̅�𝑖𝛿𝛿2 + 𝜈𝜈𝑖𝑖𝑖𝑖2) (14)

and obtain generalized residuals from the estimated coefficients as:

�̂�𝜇𝑖𝑖𝑖𝑖2 = ℎ𝑖𝑖𝑖𝑖2�𝑦𝑦𝑖𝑖𝑖𝑖2, �̂�𝜅𝑖𝑖2 + 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋�2 + 𝑧𝑧�̅�𝑖𝛿𝛿2� = 𝑦𝑦𝑖𝑖𝑖𝑖2𝜆𝜆��̂�𝜅𝑖𝑖2 + 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋�2 + 𝑧𝑧�̅�𝑖�̂�𝛿2�

−(1 − 𝑦𝑦𝑖𝑖𝑖𝑖2)𝜆𝜆� −�̂�𝜅𝑖𝑖2 − 𝑧𝑧𝑖𝑖𝑖𝑖𝜋𝜋�2 − 𝑧𝑧�̅�𝑖𝛿𝛿2� (15)

In the second stage, we estimate the structural equation (12) for the household labor

force participation rate by adding generalized residuals obtained in the first stage as additional

explanatory variables. The empirical model is given by:

𝑦𝑦𝑖𝑖𝑖𝑖1 = 𝑥𝑥𝑖𝑖𝑖𝑖1𝛽𝛽0 + 𝑦𝑦𝑖𝑖𝑖𝑖2𝑥𝑥𝑖𝑖𝑖𝑖1𝛾𝛾1 + 𝑧𝑧�̅�𝑖𝜌𝜌0 + 𝑦𝑦𝑖𝑖𝑖𝑖2𝑧𝑧�̅�𝑖𝜌𝜌1 + 𝜉𝜉0�̂�𝜇𝑖𝑖𝑖𝑖2 + 𝜉𝜉1𝑦𝑦𝑖𝑖𝑖𝑖2�̂�𝜇𝑖𝑖𝑖𝑖2 + 𝜔𝜔𝑖𝑖𝑖𝑖 (16)

𝑖𝑖 = 1,⋯ ,𝑁𝑁, 𝑡𝑡 = 1,⋯ ,𝑇𝑇

21

In our empirical model, 𝑥𝑥𝑖𝑖𝑖𝑖 contains only exogenous variables. Thus, we estimate the model

(16) by pooled OLS for all N and T. As the model (16) has a generator regressor problem, we

bootstrap to obtain standard errors. The joint significance of (𝜉𝜉0�̂�𝜇𝑖𝑖𝑖𝑖2, 𝜉𝜉1𝑦𝑦𝑖𝑖𝑖𝑖2�̂�𝜇𝑖𝑖𝑖𝑖2) in (16) implies

the endogeneity of self-selection of the migration or remittances. We test this by the Wald

statistic with two-degrees of freedom.

5. Data

We utilize data from the World Bank’s Listening to Tajikistan (L2TJK) survey. This

phone-based high frequency panel survey monitors a variety of indicators including migration,

income and employment, the wellbeing and life satisfaction of households, and access to water

and electricity services. A sample of 800 households was drawn from a nationally representative

face-to face survey of 3000 households in Tajikistan conducted in the spring of 2015. Our

analysis covers 32 rounds of the L2TJK survey from May 2015 to November 2017 in which

households were initially interviewed at 10-day intervals, changing to two-week intervals after

the sixth wave of the data collection. Households who refused to participate were replaced with

households from the same primary sampling unit (PSU). The Japan International Cooperation

Agency Research Institute (JICA-RI) joined the World Bank to contribute to the financing of the

L2TJK survey from the 31st round and added additional questions that cover special issues

about migration and remittances to the survey questionnaire form.

For each round of the L2TJK survey, we obtained information on household

characteristics such as the number of employed and unemployed, the number of children under

18 years old, the number of the elderly above 60 years old, and the household head’s age, gender,

and educational level. While this high-frequency panel dataset is unique, it is not free of

limitations. Due to the nature of the data collection method that makes the high frequency

22

household panel data possible, most indicators are collected at household level. Therefore,

individual level data for health, education, and labor market participation are not observed,

leaving us unable to estimate more detailed results by gender and age. Despite this limitation, the

high frequency panel dataset improves the efficiency of econometric estimates and allows us

greater capacity to capture variations in household behavior regarding labor supply.

Table 1. Summary statistics of variables of interest by migrant status

Total sample Non-migrants Migrants

Mean Std. Err. Mean Std. Err. Mean Std. Err. Remittance receiving 0.098 0.002 (omitted) 0.304 0.005 Migrant sending 0.323 0.003 (omitted) 1.000 0.000 Labor force participation rate 0.209 0.001 0.237 0.002 0.150 0.002 Economic activity rate 0.337 0.002 0.374 0.002 0.260 0.003 Household size 6.703 0.019 6.234 0.020 7.687 0.036 Number of the elderly aged above 60 0.432 0.004 0.402 0.005 0.496 0.008 Number of children below 18 2.634 0.011 2.544 0.013 2.824 0.021 Number of disabled 0.152 0.003 0.165 0.003 0.126 0.004 Household head's age 53.55 0.085 52.558 0.107 55.632 0.132 Male headed 0.792 0.003 0.807 0.003 0.763 0.005 Female headed 0.208 0.003 0.193 0.003 0.237 0.005 Head's marital status: Married 0.784 0.003 0.782 0.003 0.789 0.004

Divorced 0.035 0.001 0.041 0.002 0.021 0.002

Widowed 0.152 0.002 0.141 0.003 0.175 0.004

Separated 0.009 0.001 0.011 0.001 0.004 0.001

Not registered 0.012 0.001 0.015 0.001 0.007 0.001

Single 0.008 0.001 0.010 0.001 0.004 0.001 Head's education level, years 10.891 0.018 10.995 0.022 10.673 0.029 Number of observations 25,550 17,303 8,247

Source: Authors’ computation based on the L2TJK.

23

Table 2. Summary statistics of variables of interest by remittance status

Non-remitters Remitters

Mean Std. Err. Mean Std. Err. Remittance receiving (omitted) 1.000 0.000 Migrant sending 0.249 0.003 1.000 0.000 Labor force participation rate 0.216 0.001 0.142 0.003 Economic activity rate 0.348 0.002 0.241 0.005 Household size 6.571 0.019 7.914 0.072 Number of the elderly aged above 60 0.421 0.004 0.538 0.014 Number of children below 18 2.597 0.011 2.980 0.039 Number of disabled 0.155 0.003 0.128 0.008 Household head's age 53.282 0.090 56.014 0.242 Male headed 0.796 0.003 0.759 0.009 Female headed 0.204 0.003 0.241 0.009 Head's marital status: Married 0.784 0.003 0.789 0.008

Divorced 0.036 0.001 0.021 0.003

Widowed 0.150 0.002 0.175 0.008

Separated 0.009 0.001 0.004 0.001

Not registered 0.013 0.001 0.008 0.002

Single 0.008 0.001 0.004 0.001 Head's education level, years 10.928 0.019 10.554 0.057 Number of observations 23,044 2,506

Source: Authors’ computation based on the L2TJK.

Based on the main interview data, we constructed variables for migration and remittance

status, labor force participation rate, and economic activity rate of the household members

remaining in Tajikistan (see Tables 1 and 2). The remittance receiving status variable was

constructed as a dummy variable to account for the receipt of remittances by a household from

its migrant members during a survey round. Similarly, the migration variable is a binary variable

that takes on the value of 1 if a household has sent at least one member abroad. These two

dummy variables serve as regime-switchers in our structural model of labor force participation

and the rate of economic activity. We use both remittances and migration as regime-switchers

because many Tajik migrants are short-term seasonal migrants and do not remit but bring the

money home with them. Later, we also show our results with continuous variables for remittance

amount and number of migrants as a robustness check.

The data show that 32.3 percent of the total households have at least one migrant

member; however, only 9.8 percent receive remittances from their migrant members. The

summary statistics also suggest that households with migrants have lower labor force

24

participation and economic activity rates than non-migrant sending households. The indicators

are even lower for remittance receiving households than migrant sending households.

In total, we have two dependent variables and two regime-switching variables, leading

to four endogenous switching model specifications. Each endogenous switching model has one

component that is the endogenous regime-switching variable. Thus, we need at least one

instrument in the first stage probit model that is not included in the structural model. We

employed two instruments: monthly wage rates at the migration destination and the number of

migrants at the primary sampling unit (PSU).

The instrumental variables (IVs) were chosen on the basis of the theoretical and

empirical literature of migration. The Harris-Todaro (1970) model predicts that the most

important determinants of migration are the wage differentials between home and destination.

However, it is difficult to construct a variable for wage differentials because wage data can only

be observed for an individual in either home or foreign countries, but not simultaneously. Thus,

for a practical reason, we use wage data from destinations, assuming that wage rates are higher

in foreign countries than at home. For these wage data, we compiled information on monthly

wages in local currencies from the major destination countries, including Russia, Kazakhstan,

China, Turkey, South Korea, the United States, and Ukraine from the corresponding months and

quarters of 2015 to 2017 that match with the L2TJK data. Monthly wage data for Russia came

from the Russian Federation Federal State Statistics Service. Data for Kazakhstan came from the

Ministry of National Economy of the Republic of Kazakhstan Committee on Statistics. Hourly

wage data for Turkey came from the Turkish Statistical Institute and data on the quarterly wage

of migrant laborers in China came from the National Bureau of Statistics of China. To enable

comparison, the amounts in local currencies were converted into US Dollars using historical

exchange rates from the United States Department of the Treasury. For non-migrant sending

households, we took the average wage of the all destination countries except the United States

and South Korea as a negligible share of Tajik migrants work in these countries.

25

Table 3. Summary statistics of IVs

Instrumental Variables Mean Std. dev. Min Max Number of migrants in PSU 2.29 1.94 0 11 Wage rate at destination 3995.97 953.71 2578.44 13501.08 Number of observations 25,550

Source: Authors’ computations.

Finally, the New Economics of Labor Migration (NELM) theory emphasizes the

importance of the network effect of migration as a key determinant of labor migration.

Particularly, it argues that using personal networks in the destination could reduce migration

costs and thereby promote more migration. Past empirical studies that test the NELM hypothesis

often use the number of migrants in the community or the presence of return migrants as proxies

for a migration network. Thus, we follow past literature and use the number of migrants in the

community as a proxy for the migration network. In this paper, we use the PSU as a community

as it is the smallest unit that the sample could be drawn from. The surveyed sample has 150 PSUs,

each containing 5-10 households. We constructed the instrumental variable by adding up

migrant households in the PSU. Table 3 summarizes the two instrumental variables.

6. Results and discussion

We applied the approach described in Section 4 to four separate cases with combinations of two

endogenous regime-switching variables and two response outcome variables. We take a

household’s migration and remittance receiving statuses as regime-switching variables, and

household’s labor force participation rate and economic activity rate as response outcome

variables. In the first stage, we estimate two pooled probit models of migration and remittances

status respectively for all households and rounds. Since each structural model has one

26

endogenous component, we need to include at least one instrumental variable in the first stage

probit models for robust estimations. As described in the previous section, we have two

instruments: wage rates in destinations and the number of migrants in the PSU. By selecting

wage rates and the number of migrants as instruments, we assume that they have no direct effect

on household labor supply decisions once we control for household migration or remittance

decisions. Table 4 reports the first stage coefficient estimates for the probit models.

As required by the two-step estimation procedure, the probit models in Table 4 also

include time averages of all explanatory variables except for the time-invariant variables, and

regional and time dummy variables as they are perfectly collinear with the constant term. The

results in Table 4 suggest that most estimated coefficients are statistically significant and the

directions of the effect of household characteristics on migration and remittance decisions are

consistent across the models, with larger magnitudes for the migration decision in general.

The results suggest that larger households with a married, older, and female head tend to

send migrants abroad. Households with more elderly members aged over 60 tend to send

migrants and receive remittances, whereas households with handicapped members are less likely

to send migrants and receive remittances. Having more children below 18 years of age

significantly reduces the probability of having a migrant household member. Finally, the

educational level of household head is negatively related to migration and remittances.

The explanatory variables that serve as instruments in our endogenous switching model

are statistically significant in both probit models. As the number of migrants in the PSU

increases, the probability of sending migrants and receiving remittances increases. This is

consistent with the migration literature that suggests that the migration network is an important

determinant of the decision to migrate. While we did not consider the wage differentials as

predicted by the Todaro model, our results suggest that higher wages at the migration

destinations attract migrants and increase the probability of sending remittances. We exclude

these two explanatory variables (the number of migrants in the PSU and the wage rate in the

27

destination) from the structural equations of the household labor supply to exploit them as

instruments. Thus, it is assumed that the number of migrants in the PSU and the wage rate in a

destination have no direct effect on labor supply decisions at home.

Table 4. First stage coefficient estimates: Determinants of the migration and remittance decisions Remittances Migrants Household size 0.065 0.147

(5.30)*** (12.33)*** Number of elderly (60+) 0.004 0.049

(0.100) (1.290) Number of children (<18) -0.044 -0.141

(-2.26)** (-7.82)*** Number of disabled -0.085 -0.145

(-2.86)*** (-5.53)*** Head's age 0.041 0.057

(4.80)*** (7.57)*** Head's age squared 0.000 -0.001

(-4.23)*** (-7.00)*** Female headed 0.245 0.274

(5.98)*** (6.99)*** Head's marital status (Reference: Married)

Divorced -0.316 -0.385

(-3.80)*** (-5.27)***

Widowed -0.129 -0.064

(-2.61)*** (-1.37)

Separated -0.461 -0.958

(-2.79)*** (-5.99)***

Not registered marriage -0.239 -0.521

(-1.94)* (-4.70)***

Single -0.520 -0.896

(-2.70)*** (-4.37)*** Head's education in years -0.027 -0.014

(-2.84)*** (-1.70)* Time averaged variables Yes Yes Regional dummies Yes Yes Time dummies Yes Yes Instruments

Number of migrants in the PSU 0.155 0.279

(18.36)*** (31.12)***

Wage rate at destination 0.0002 0.005

(16.32)*** (63.07)*** Constant -3.892 -20.77

(-17.94)*** (-61.07)*** Number of observations 25,550 25,550 Source: Authors’ estimates. Notes: ***, ** and * indicate statistical significance at 1 percent, 5 percent, and 10 percent respectively. t-statistics are in parentheses. Time variables are survey rounds.

28

In the second stage of the estimation procedure, the structural equations of the

household labor supply are augmented with the generalized residuals obtained from the

estimates of the first-stage probit models to correct for the endogeneity and the self-selection

bias. Table 5 reports on the parameter estimates of the household labor supply as measured by

household labor force participation and economic activity rates.

Table 5. Second-stage coefficient estimates: the determinants of labor force participation and economic activity rates Dependent variable Labor supply Labor supply Economically active Economically active Regime-switcher variable Remittance Migrant Remittance Migrant Remittances/Migrants -0.112 -0.080 -0.198 -0.097

(-2.55)** (-3.56)*** (-3.36)*** (-3.69)*** Household size -0.004 -0.007 -0.001 -0.004

(-2.63)*** (-3.90)*** (-0.82) (-1.56) Number of elderly (60+) -0.022 -0.028 -0.040 -0.049

(-4.86)*** (-5.21)*** (-6.30)*** (-6.64)*** Number of children (<18) -0.023 -0.024 -0.042 -0.044

(-10.59)*** (-10.44)*** (-14.88)*** (-13.29)*** Number of disabled -0.038 -0.047 -0.043 -0.053

(-14.83)*** (-17.00)*** (-13.14)*** (-15.29)*** Head's age 0.002 0.002 0.005 0.005

(1.93)* -1.620 (4.93)*** (4.00)*** Head's age squared 0.000 0.000 0.000 0.000

(-1.51) (-1.18) (-4.10)*** (-3.46)*** Female headed -0.020 -0.027 -0.024 -0.039

(-4.71)*** (-4.37)*** (-4.64)*** (-5.42)*** Head's marital status (Reference: Married)

Divorced 0.097 0.096 0.092 0.090

(9.18)*** (7.93)*** (10.30)*** (7.86)*** Widowed 0.000 0.005 -0.021 -0.009

0.000 -0.740 (-3.22)*** (-1.04) Separated 0.222 0.247 0.206 0.211

(9.47)*** (10.80)*** (9.35)*** (8.45)*** Not registered

marriage -0.036 -0.039 -0.048 -0.045

(-3.73)*** (-4.28)*** (-3.94)*** (-3.16)*** Single 0.058 0.051 -0.015 -0.027

(2.82)*** (2.08)** (-0.73) (-1.06) Head's education in years 0.005 0.005 -0.003 -0.003

(4.61)*** (4.67)*** (-2.40)** (-2.19)** Time averaged variables Yes Yes Yes Yes Regions Yes Yes Yes Yes Rounds Yes Yes Yes Yes Interactions with remittances/migrants

Household size 0.002 0.008 0.007 0.008

-0.670 (2.92)*** -1.190 (2.34)** Number of elderly

(60+) 0.026 0.019 0.036 0.033

(1.92)* (2.37)** (1.94)* (2.93)***

29

Number of children (<18) 0.012 0.005 0.007 0.006

(2.07)** -1.400 -0.870 -1.190 Number of disabled 0.034 0.034 0.065 0.059

(3.68)*** (6.19)*** (4.99)*** (7.42)*** Head's age 0.009 0.005 0.012 0.006

(3.02)*** (2.51)** (3.28)*** (2.56)** Head's age squared 0.000 0.000 0.000 0.000

(-3.28)*** (-2.58)*** (-3.27)*** (-2.29)**

Source: Authors’ estimates. Notes: ***, ** and * indicate statistical significance at 1%, 5%, and 10% respectively. t-statistics are in parentheses. Time variables are survey rounds.

All the regressions reported in Table 5 contain full sets of regional and time dummy

variables, time-averages of time-variant variables, and interactions of all variables with the

dummies for whether the household has a migrant member or whether the household received

remittances from its migrant members respectively. All continuous variables were de-meaned

before being interacted with the regime switching dummies. Therefore, the estimated coefficient

on the regime switching dummy variables can be meaningfully interpreted as average treatment

Female headed -0.017 0.017 0.009 0.048

(-1.35) (1.83)* -0.560 (4.40)*** Divorced -0.068 -0.046 -0.062 -0.047

(-2.64)*** (-2.23)** (-1.66)* (-1.97)** Widowed 0.025 -0.012 0.011 -0.034

-1.470 (-0.95) -0.530 (-2.39)** Separated -0.223 -0.263 -0.120 -0.108

(-3.65)*** (-6.89)*** (-0.97) (-2.00)** Not registered marriage -0.022 0.008 -0.055 -0.023

(-0.69) -0.330 (-1.06) (-0.76) Single 0.152 0.073 0.266 0.129

(2.68)*** -1.330 (6.15)*** (2.59)*** Head's education in years -0.007 -0.003 0.000 0.004

(-2.56)** (-1.89)* -0.010 -1.510 Time averaged variables with interaction Yes Yes Yes Yes

Regions with interaction Yes Yes Yes Yes Rounds with interaction Yes Yes Yes Yes Generalized residuals from Stage 1 0.068 -0.023 0.119 -0.019

(7.85)*** (-3.32)*** (11.21)*** (-1.98)** Interacted generalized residuals -0.057 0.030 -0.082 0.032

(-3.57)*** (3.79)*** (-3.86)*** (3.11)*** Constant 0.119 0.113 0.310 0.331

(6.19)*** (5.19)*** (13.44)*** (12.09)*** Number of observations 25,550 25,444 25,550 25,444

30

effects. The remaining coefficients can be interpreted as the effect of migration/remittances on

the labor supply rates for households with given average characteristics.

To prove the validity of the endogenous switching model, we test the joint significance

of the generalized residuals terms using the Wald test with two degrees of freedom. In all models,

we reject the joint insignificance of the generalized residual terms at p=0.01 level of significance,

validating that the regime switching is endogenous.

According to the summary statistics reported in Tables 1 and 2, on average, the labor

force participation and economic activity rates are lower for migrant sending and remittance

receiving households. This observation is supported by our estimates of the structural equations

as presented in Table 5. The average treatment effect coefficients of the migration and

remittances are all negative and highly statistically significant, implying that migrant sending

and remittance receiving households have lower labor supply rates. In terms of magnitude, the

negative impact of remittances is larger than that of migrants. Our results show that the presence

of a migrant member reduces the labor force participation rate of the remaining household

members by 8 percentage points, while the receipt of remittances reduces it by 11 percentage

points. Furthermore, the response of the economic activity rate is larger than that of the

participation rate. Having a migrant member reduces a household’s economic activity rate by 9.7

percentage points compared to the 19.8 percentage point reduction due to the receipt of

remittances.

Our results are consistent with past research in the Tajikistan context. Justino and

Shemyakina (2012) also observe negative impacts from migration and remittances on the labor

force participation of both men and women, although they do not correct for the endogeneity and

selection bias of migration and remittances. In terms of the degree of the impact, their findings

show that receiving remittances and having a migrant member reduces the labor force

participation rate of men by 8 percent and 1 percent respectively. The impacts for women are 5

percent and 2 percent respectively.

31

The effects of other determinants of household labor supply rates depend on whether the

household has a migrant member and receives remittances. For non-migrant and non-remittance

receiving households, large households with more young and old dependents, having

handicapped members and with a female head are likely to have lower labor force participation

rates, whereas the household head’s age and education may increase their participation rates.

The effects on the economic activity rates for non-migrant and non-remittance receiver

households also follow the same pattern.

For migrant supplying and remittance receiving households, the effects of the

determinants should be discussed in conjunction with the results of the first stage probit model.

Because migrant households tend to have fewer children aged below 18 and a lower number of

disabled members, having these dependents increases their labor force participation rates,

perhaps due to an increased need for income to take care of them. Also, older female-headed

migrant households are more likely to participate in the labor market. The educational level of

the household heads of migrant households is positively related to the economic activity rate, but

negatively with the participation rate, indicating some degree of mismatch in the labor market.

7. Robustness check with alternative specifications

We conduct several robustness checks to confirm the validity of our main results. We estimated

three panel data models with both binary and continuous endogenous variables. For the

continuous endogenous variables, we use the amount of remittances received and the number of

migrant household members. The models that we estimated include an Anderson-Hsiao (1981)

type dynamic panel model, a standard fixed-effect model, and a lagged-dependent variable

(ANCOVA) model. The results are consistent with our main results, although the magnitude of

the estimates is slightly lower except for the impact of the economic activity rate coefficients on

the remittance status. For both continuous endogenous explanatory variables, the negative

32

effects are significantly and consistently observed, which further supports our findings in

relation to the regime-switching regression. Generally the magnitude of the impact of

remittances is larger than that of migration, implying that remittances have much larger income

effect that discourages labor market participation on non-migrant household members.

Our benchmark panel model for robustness check is as follows:

𝑦𝑦𝑖𝑖𝑖𝑖 = 𝛽𝛽𝐷𝐷𝑖𝑖𝑖𝑖 + 𝛾𝛾𝑋𝑋𝑖𝑖𝑖𝑖 + 𝜃𝜃𝑦𝑦𝑖𝑖,𝑖𝑖−1 + 𝜇𝜇𝑖𝑖 + 𝜖𝜖𝑖𝑖𝑖𝑖 (17)

Where: 𝑦𝑦𝑖𝑖𝑖𝑖 is the outcome variables such as labor supply and economically active labor

participation; and 𝐷𝐷𝑖𝑖𝑖𝑖 are key explanatory variables of interest, such as the status of remittance

and migration. In addition to the discrete dummy variables we use in the main text, 𝐷𝐷𝑖𝑖𝑖𝑖 also

includes the continuous treatment variables, amount of remittance and number of migrants from

each household. 𝜇𝜇𝑖𝑖 represents household 𝑖𝑖’s time-invariant unobserved characteristics, and 𝜖𝜖𝑖𝑖𝑖𝑖

is a mean-zero idiosyncratic shock.

If there is a serial correlation, meaning that the past idiosyncratic shock, 𝜖𝜖𝑖𝑖,𝑖𝑖−1, is

correlated with the current outcome, 𝑦𝑦𝑖𝑖𝑖𝑖, it is well known that the standard fixed effect model

will not give a consistent estimate. For the variables related to labor supply, it is highly probable

that past shocks can affect the present decision. This is the reason for including the lagged

dependent variable, 𝑦𝑦𝑖𝑖,𝑖𝑖−1, in the right-hand-side of (17).

In practice, we use the first differenced equation of (17) so that we can eliminate 𝜇𝜇𝑖𝑖:

𝛥𝛥𝑦𝑦𝑖𝑖𝑖𝑖 = 𝛽𝛽𝛥𝛥𝐷𝐷𝑖𝑖𝑖𝑖 + 𝛾𝛾𝛥𝛥𝑋𝑋𝑖𝑖𝑖𝑖 + 𝜃𝜃𝛥𝛥𝑦𝑦𝑖𝑖,𝑖𝑖−1 + 𝛥𝛥𝜖𝜖𝑖𝑖𝑖𝑖 (18)

In addition to our key identification challenge, the endogeneity of 𝛥𝛥𝐷𝐷𝑖𝑖𝑖𝑖 , 𝛥𝛥𝜖𝜖𝑖𝑖𝑖𝑖 is obviously

correlated with 𝛥𝛥𝑦𝑦𝑖𝑖,𝑖𝑖−1 as both have the common unobservable 𝜖𝜖𝑖𝑖,𝑖𝑖−1 . A conventional

approach to deal with this endogeneity is to instrument 𝛥𝛥𝑦𝑦𝑖𝑖,𝑖𝑖−1 with 𝑦𝑦𝑖𝑖,𝑖𝑖−2 as suggested by

33

Anderson and Hsiao (1981). Thus, in (18) we instrument two endogenous variables 𝛥𝛥𝐷𝐷𝑖𝑖𝑖𝑖 and

𝛥𝛥𝑦𝑦𝑖𝑖,𝑖𝑖−1 by the first-differenced PSU level number of migrants, the first-differenced wage rate at

the destination, and 𝑦𝑦𝑖𝑖,𝑖𝑖−2.

The regression results for equation (18) are summarized in Table 6. The results are

similar to our main results, both in terms of the sign and the magnitude of the estimates.

Furthermore, the order of the magnitude among different combinations of

dependent-explanatory variables is also consistent with that of our main results. Columns (1) and

(5) of Table 6 report the results when 𝐷𝐷𝑖𝑖𝑖𝑖 is the dummy of receiving remittances. Receipt of

remittances reduces the labor force participation rate by 9.2 percentage points and the economic

activity rate by 27.1 percentage points. Columns (2) and (6) show the results when the

remittance dummy is replaced by the log of the remittance amount. The results are consistent

with the results using the remittance dummy, and they suggest that doubling remittances reduces

the labor force participation rate by 1.3 percentage points and the economic activity rate by 3.9

percentage points. The effect of the presence of migrant(s) is displayed in columns (3) and (7).

The labor force participation rate declines by 3.2 percentage points if there is at least one migrant

in the household. The effect is even larger for the economic activity ratio, with an 8.6 percentage

points decline. As the continuous counterpart of the dummy of the presence of migrants, we

estimate the effect of the number of migrants as reported in columns (4) and (8). Adding one

migrant will reduce the labor force participation by 1.8 percentage points while reducing the

economic activity rate by 5.2 percentage points.

This Anderson-Hsiao estimator, though being widely used, requires the assumption

that 𝜖𝜖𝑖𝑖𝑖𝑖 is not serially correlated, which could be too strong in some cases. Angrist and Pischke

(2009) suggest testing the robustness with two alternative specifications, the fixed-effect

estimation, and the lagged-dependent estimation (sometimes called “ANCOVA”) which can

jointly give a nice “bracket” (the upper-bound and the lower-bound) for the estimate. The fixed

34

effect estimation ignores the lagged dependent variable from equation (17). We take the first

difference and estimate:

𝛥𝛥𝑦𝑦𝑖𝑖𝑖𝑖 = 𝛽𝛽𝐹𝐹𝐹𝐹𝛥𝛥𝐷𝐷𝑖𝑖𝑖𝑖 + 𝛾𝛾𝐹𝐹𝐹𝐹𝛥𝛥𝑋𝑋𝑖𝑖𝑖𝑖 + 𝛥𝛥𝜖𝜖𝑖𝑖𝑖𝑖 (19)

Instead, in the ANCOVA model, we drop 𝜇𝜇𝑖𝑖 from (17) and estimate:

𝑦𝑦𝑖𝑖𝑖𝑖 = 𝛼𝛼 + 𝛽𝛽𝐴𝐴𝐴𝐴𝐷𝐷𝑖𝑖𝑖𝑖 + 𝛾𝛾𝐴𝐴𝐴𝐴𝑋𝑋𝑖𝑖𝑖𝑖 + 𝜃𝜃𝐴𝐴𝐴𝐴𝑦𝑦𝑖𝑖,𝑖𝑖−1 + 𝜖𝜖𝑖𝑖𝑖𝑖 (20)

The results of the estimation of (19) and (20) are reported in Tables 7 and 8, respectively.

For all the estimates, the results are similar to the corresponding values in Table 6. Estimates in

the columns (1), (3), (5), and (7) of each table are qualitatively the same as the corresponding

main results which appear in columns 1 to 4 of Table 4, respectively. Table 9 summarizes the

coefficient across different specifications, so that the reader can easily compare the results. From

all of the estimation results, it is highly probable that our estimates of the impact of remittances

and migration on the labor supply of left-behinds, which is significantly negative and sizable, are

stable across different empirical specifications.

35

Table 6. Anderson-Hsiao estimation (Equation 18) (1) (2) (3) (4) (5) (6) (7) (8)

VARIABLES 𝛥𝛥Labor Supply

𝛥𝛥Labor Supply

𝛥𝛥Labor Supply

𝛥𝛥Labor Supply

𝛥𝛥Economically Active

𝛥𝛥Economically Active

𝛥𝛥Economically Active

𝛥𝛥Economically Active

𝛥𝛥Remittance -0.0921**

-0.271***

(0.0460)

(0.0610) 𝛥𝛥Remittance

Amount

-0.0134**

-0.0396***

(0.00676)

(0.00898)

𝛥𝛥Migrants

-0.0322**

-0.0860***

(0.0138)

(0.0170)

𝛥𝛥Number of migrants

-0.0178**

-0.0518***

(0.00862)

(0.0107)

𝛥𝛥Labor supply (t-1) 0.106*** 0.108*** 0.105*** 0.105***

(0.0233) (0.0234) (0.0233) (0.0233)

𝛥𝛥Econ. Active (t-1)

0.0888*** 0.0934*** 0.0945*** 0.0942***

(0.0200) (0.0199) (0.0191) (0.0191)

Observations 22,611 22,611 22,611 22,611 22,611 22,611 22,611 22,611 R-squared -0.112 -0.115 -0.092 -0.091 -0.223 -0.224 -0.084 -0.083 Controls YES YES YES YES YES YES YES YES Regions YES YES YES YES YES YES YES YES Rounds YES YES YES YES YES YES YES YES Robust standard errors are in parentheses.

*** p<0.01, ** p<0.05, * p<0.1 Dependent variables are the labor force participation rate in level for the columns (1) to (4), and the economic activity rate for the column (5) to (8).

“Remittance Amount” is the log of (1 + remittance amount) to included households without remittance received into the sample. “Number of migrants” is the raw number of migrants in the household. The first stage results can be provided upon request.

36

Table 7. First-differenced equation model (Equation 19)

(1) (2) (3) (4) (5) (6) (7) (8)

VARIABLES Labor Supply

Labor Supply

Labor Supply

Labor Supply

Economically Active

Economically Active

Economically Active

Economically Active

Remittance -0.0803*

-0.257***

(0.0439)

(0.0592) Remittance

Amount

-0.0117*

-0.0376***

(0.00643)

(0.00867)

Migrants

-0.0268**

-0.0772***

(0.0126)

(0.0158)

# of migrants

-0.0145*

-0.0469***

(0.00797)

(0.0100)

Observations 24,007 24,007 24,007 24,007 24,007 24,007 24,007 24,007 R-squared -0.002 -0.003 0.012 0.012 -0.116 -0.112 0.011 0.011 Number of hhid 1,346 1,346 1,346 1,346 1,346 1,346 1,346 1,346 Controls YES YES YES YES YES YES YES YES Rounds YES YES YES YES YES YES YES YES Robust standard errors are in parentheses.

*** p<0.01, ** p<0.05, * p<0.1 The first stage results can be provided upon request.

37

Table 8. ANCOVA model (Equation 20) (1) (2) (3) (4) (5) (6) (7) (8)

VARIABLES

Labor

Supply

Labor

Supply

Labor

Supply

Labor

Supply

Economically

Active

Economically

Active

Economically

Active

Economically

Active

Remittance -0.129***

-0.202***

(0.0151)

(0.0197)

Remittance

Amount

-0.0174***

-0.0268***

(0.00198)

(0.00260)

Migrants

-0.0410***

-0.0628***

(0.00455)

(0.00583)

# of migrants

-0.0278***

-0.0433***

(0.00315)

(0.00401)

Observations 24,007 24,007 24,007 24,007 24,007 24,007 24,007 24,007

R-squared 0.396 0.399 0.419 0.417 0.400 0.407 0.437 0.436

Controls YES YES YES YES YES YES YES YES

Regions YES YES YES YES YES YES YES YES

Rounds YES YES YES YES YES YES YES YES Robust standard errors are in parentheses.

*** p<0.01, ** p<0.05, * p<0.1 The first stage results can be provided upon request.

38

Table 9. Results summary across different specifications (1) (2) (3) (4) Dependent Variable

Remittance/migration Variables

Regime-Switching model Anderson-Hsiao Fixed-Effect ANCOVA

Labor Supply Remittance -0.112** -0.0921** -0.0803* -0.129***

Remittance Amount N.A. -0.0134** -0.0117* -0.0174***

Migrants -0.080*** -0.0322** -0.0268** -0.0410***

# of migrants N.A. -0.0178** -0.0145* -0.0278***

Economically Active

Remittance -0.198*** -0.271*** -0.257*** -0.202***

Remittance Amount N.A. -0.0396*** -0.0376*** -0.0268***

Migrants -0.097*** -0.0860*** -0.0772*** -0.0628***

# of migrants N.A. -0.0518*** -0.0469*** -0.0433***

*** p<0.01, ** p<0.05, * p<0.1

39

8. Conclusions

Out-migration has increased rapidly in Tajikistan and is likely to rise further in response to the

economic incentives offered by neighboring countries, especially Russia. Private remittances

from migrant workers contribute to Tajikistan’s economy excessively, at its highest in 2008

making up almost 50 percent of its GDP. While remittance receipts in Tajikistan have recently

been in decline because of migrants returning from Russia and the economic slowdown in

Russia, migration remains to be a lucrative and preferred choice of occupation for many Tajiks.

This paper has explored the labor market impact of overseas out-migration and

remittances in Tajikistan using the unique high frequency household panel data, L2TJK. The

analysis covered 32 rounds of the L2TJK survey collected between 2015 and 2017. To consider

the possibility of endogeneity and selection bias in the migration and remittances decisions, we

employed a control function approach to endogenous switching regression for panel data

developed by Murtazashvili and Wooldridge (2016). The advantage of applying the control

function approach to endogenous switching regression is that it is less restrictive as it allows

serial correlation in the error term as well as heterogeneities to be correlated with time-varying

explanatory variables. This approach is less computationally expensive than the alternative full

information maximum likelihood approaches.

Our results show that having a migrant member or receiving overseas remittances can

reduce labor force participation and economic activity rates of the remaining household

members. The remaining household members’ participation in the labor market is more

responsive to remittances than to migration. This result is in line with past empirical studies of

the type in Tajikistan and other countries, as well as previous theoretical constructs. Our findings

suggest that migration and remittances raise the reservation wages of members left in the

household according to the labor-leisure theory that states that remittances, when considered as

40

non-labor income, can decrease the propensity of non-migrant household members to participate

in the labor market.

The results of the study add to the debate on how remittances and migration can

ultimately impact on development. There are several channels through which remittances and

migration can affect the wellbeing of households – as a buffer to shocks, by increasing the

per-capita income of households, through improvements in access to education, health, and

other well-being indicators of household members, among others. This paper looked at one

channel, one that is important in assessing long-term economic growth.

The policy implications of the results depend on what migrant households are doing

instead of working. If they are taking on unpaid household work previously borne by migrant

members, it could imply a need to improve the wage labor market. Detailed information about

the time use of household members is not available in Tajikistan and limits the possibility of

performing more detailed analysis disaggregated by age and gender. This paper shows that

collecting such data can improve our collective knowledge of the impact of migration and

remittances on domestic labor supply. While the results show that migration and remittances

may have a negative impact on the labor market supply of the household members left behind,

the results do not deny the possibility of remittances and migration having a positive impact on

other outcomes. Thus, this paper encourages further study to piece together a more complete

picture that would allow us to suggest better policy responses on how to channel such

remittances into development.

41

References:

Anderson, T. W., and Hsiao Cheng. 1981. “Estimation of dynamic models with error components.” Journal of the American Statistical Association 106 (6): 589-606.

Angrist, Joshua D., and Jorn-Steffen Pischke. 2009. Mostly harmless Econometrics: An Empiricist’s Companion. New York: Princeton University Press.

Abdulloev, Ilhom. 2013. “Impact of Migration on Job Satisfaction, Professional Education and the Informal Sector.” PhD Diss., Rutgers University.

Acosta, Pablo. 2007. “Entrepreneurship, Labor Markets and International Remittances: Evidence from El Salvador.” In International Migration Policy and Economic Development, edited by Çaglar Özden and Maurice Schiff. Washington, DC: International Bank for Reconstruction and Development and the World Bank.

Acosta, Pablo, Cesar Calderon, Pablo Fajnzylber, and J. Humberto López. 2008. “Do Remittances Lower Poverty Levels in Latin America?” In Remittances and Development Lessons from Latin America, edited by Pablo Fajnzylber and J. Humberto López. Washington, DC: World Bank.

Acosta, Pablo, Emmanuel K. K. Lartey, and Federico S. Mandelman. 2009. “Remittances and the Dutch Disease.” Journal of International Economics 79 (1): 102-16.

Adams Jr., Richard H. 2011. “Evaluating the Economic Impact of International Remittances On Developing Countries Using Household Surveys: A Literature Review.” Journal of Development Studies 47 (6): 809-28.

Binzel, Christine, and Ragui Assaad. 2011. “Egyptian men working abroad: Labor Supply Responses by the Women Left Behind.” Labour Economics 18 (S1): 98-114.

Brown, Richard, Olimoba Saodat, and Boboev Mukhammady. 2008. “Country Report on Remittances of International Migrants in Tajikistan.” In Study on International Migrants’ Remittances in Central Asia and South Caucasus. Manilla: Asian Development Bank.

Cabegin, Emily Christi A. 2006. “The Effect of Filipino Overseas Migration on the Nonmigrant.” IZA Discussion Papers 2240.

Calero, Carla, Arjun S. Bedi, and Robert Sparrow. 2009. “Remittances, Liquidity Constraints and Human Capital Investments in Ecuador.” World Development 37 (6): 1143-54.

Chami, Ralph, Connel Fullenkamp, and Samir Jahjah. 2005. “Are Immigrant Remittance Flows a Source of Capital for Development?” IMF Staff Papers 52 (1): 55-81.

Chami, Ralph, Dalia Hakura, and Peter Montiel. 2012. “Do Worker Remittances Reduce Output Volatility in Developing Countries?” Journal of Globalization and Development 3 (1): 1-25.

Cox-Edwards, Alejandra, and Eduardo Rodriguez-Oreggia. 2009. “Remittances and Labor Force Participation in Mexico.” World Development 37 (5): 1004-14.

Démurger, Sylvie. 2015. “Migration and Families Left Behind.” IZA World of Labor 144. doi: 10.15185/izawol.144.

Erlich, Aaron. 2006. Tajikistan: From Refugee Sender to Labor Exporter. Migration Policy Institute. https://www.migrationpolicy.org/article/tajikistan-refugee-sender-labor-exporter.

European Union. 2010. Labour Market Tajikistan. https://www.etf.europa.eu/sites/default/files/m/17405B3004DC50AAC12578DA003C0EE1_Labour%20market%20review_Tajikistan.pdf.

Fayissa, Bichaka, and Christian Nsiah. 2010. “The Impact of Remittances on Economic Growth and Development in Africa.” American Economist 55 (2): 92-103.

42

Funkhouser, Edward. 1992. “Mass Emigration, Remittances.” In Immigration and the Workforce: Economic Consequences, edited by George J. Borjas and Richard B. Freeman, 135-76. Chicago.

Görlich, Dennis, Toman Omar Mahmoud, and Christoph Trebesch. 2007. “Explaining Labour Market Inactivity in Migrant-Sending Families: Housework, Hammock, or Higher Education?” Kiel Institute for the World Economy Working Papers 1391.

Grigorian, David A., and Tigran A. Melkonyan. 2011. “Destined to Receive: The Impact of Remittances on Household Decisions in Armenia.” Review of Development Economics 15 (1): 139-53.

Hildebrandt, Nicole and McKenzie, David J. 2005. “The Effects of Migration on Child Health in Mexico.” World Bank Policy Research Working Papers 3573.

https://doi.org/10.1596/1813-9450-3573. International Labor Organization. 2017. ILOSTAT.

https://www.ilo.org/ilostat/faces/ilostat-home/home?_adf.ctrl-state=3uyr5nsp6_199&_afrLoop=296230194836544#!

Jansen, Dennis W., Diego E. Vacaflores, and George S. Naufal. 2012. “The Macroeconomic Consequences of Remittances.” ISRN Economics.

Justino, Patricia, and Olga N. Shemyakina. 2012. “Remittances and Labor Supply in Post-Conflict Tajikistan.” IZA Journal of Labor and Development 1 (1). doi:10.1186/2193-9020-1-8.

Kim, Namsuk. 2007. “The Impact of Remittances on Labor Supply: The Case of Jamaica.” World Bank Policy Research Working Papers 4120.

Murtazashvili, Irina, and Jeffrey M. Wooldridge. 2016. “A Control Function Approach to Estimating Switching Regression Models with Endogenous Explanatory Variables and Endogenous Switching.” Journal of Econometrics 190 (2): 252-66. doi:10.1016/j.jeconom.2015.06.014.

Ndiaye, Ameth Saloum, and Abdelkri Araar. 2017. “Migration and Remittances in Senegal: Effects on Labor Supply and and Human Capital of Households Members Left Behind.” UNU-WIDER Development Conference on Migration and Mobility. Ghana: Accra.

Olimova, Saodat. 2010. “The Impact of Labour Migration on Human Capital: The Case of Tajikistan.” Revue Européenne des Migrations Internationales 26 (3): 181-97.

Ratha, Dilip. 2013. “The Impact of Remittances on Economic Growth and Poverty Reduction.” Migration Policy Institute Policy Brief 8.

Russian Federation Federal State Statistics Services. 2016. Russia in Figures. http://www.gks.ru/. Statistical Agency of Tajikistan. 2017. Statistical agency under the President of the Republic of

Tajikistan. http://www.stat.tj/en/. Strokova, Victoria and Ajwad, Mohamed Ihsan. 2017. Jobs Diagnostic Tajikistan. Washington, DC:

World Bank. United Nations Population Division, Department of Economic and Social Affairs. 2017.

International Migrant Stock: The 2017 Revision. http://www.un.org/en/development/desa/population/migration/data/estimates2/estimates17.shtml.

World Bank. 2017. World Bank Open Data. https://data.worldbank.org. World Bank and GIZ. 2013. Tajikistan Jobs, Skills, and Migration Survey 2013. Washington, DC:

World Bank and GIZ. Yormirzoev, Mirzobobo. 2017. “Determinants of Migration Flows to Russia: Evidence from

Tajikistan.” Economics and Sociology 10 (3): 72-80.

43

Abstract (In Japanese)

要約

タジキスタンは、中央アジアにおける主要な移民送出国で、移民からの送金に対する

依存度が高い。本稿は、タジキスタンにおいて、出稼ぎ移民と送金が本国に残る家族

の労働供給に与える影響を検証している。移民・送金が家計の労働供給に与える影響

を巡っては、従来から内生性の問題が指摘されてきた。本稿では、コントロール関数

法を用いて高頻度の家計パネルデータを分析することにより、既存研究では見られな

かった方法で内生性問題の克服を試みている。分析結果は、移民の送り出しと送金の

受け取りにより、家族の留保賃金が上昇し、労働・経済活動への参加を減少されるこ

とを示唆している。この結果は、分析手法や移民・送金の定義を変えた場合でも同様

に観察され、頑健であることが示された。 キーワード：移民、海外送金、労働参加、経済活動率、内生的スイッチング、タジキ

スタン